Why is ChatGPT not working?
- 01It actually is down, and you should confirm that before anything else. OpenAI publishes incidents at status.openai.com, and its own troubleshooting articles tell you to check there — an active incident can even make chat history look empty when it is perfectly intact. Thirty seconds here saves you from resetting a password to fix somebody else's server.STATUS · PAGE
- 02The model you were using was retired. This is the cause no generic troubleshooting guide carries, and it breaks saved workflows rather than the whole app. OpenAI has retired models from ChatGPT steadily: GPT-4o, GPT-4.1, GPT-4.1 mini and o4-mini in February 2026, GPT-5.1 in March, GPT-5.2 in June, GPT-4.5 in June, and o3 on August 26, 2026. A custom GPT or bookmarked conversation pinned to a retired model stops behaving, and nothing about that reads as "retired" in the interface.MODEL · RETIRED
- 03You are signed in to a different account than you think. Google, Apple, Microsoft and email/password produce separate accounts on the same address, and your history, subscription and settings live in exactly one of them. OpenAI's own troubleshooting flags a specific trap: sign-in with Apple plus Hide My Email puts your account under a private relay address you have to look up in Apple ID settings. "My chats are gone" and "my Plus disappeared" are usually this.WRONG · ACCOUNT
- 04You have exhausted an allowance, and only one tool broke. Limits in ChatGPT are per-tool, not per-account: image generation, uploads, data analysis and individual models each have their own. When one is spent, that feature pauses until it resets while the chat carries on — which reads as "uploads are broken" rather than "I hit a limit". OpenAI shows the reset time in-product, states there is no setting to bypass it, and says support cannot reset it.ALLOWANCE
- 05Lockdown Mode is on and it is doing its job. An opt-in security setting that deliberately restricts network-enabled capabilities — live web browsing, deep research, agent mode, file downloads and some web-derived images. Everything it blocks presents as a broken feature rather than a blocked one. Personal accounts find it in Settings → Security; on a work account an admin may have set it for you, in which case there is nothing to fix at your end.LOCKDOWN · MODE
Load status.openai.com. If it is green, sign out and back in with the exact method you originally used. If your history returns, it was the account. If one specific feature is dead while chat works, it is an allowance or Lockdown Mode — not an outage.
Most "not working" reports are four different problems wearing one sentence.
We do not run uptime monitoring, and this page does not pretend to. That is a deliberate boundary: telling you ChatGPT is up right now requires infrastructure we do not have, and a page that says "it was working when we published this" is worse than no page. What we can do is separate the four failure modes that are not outages but get reported as one — because those are the ones where the fix is on your side and nobody is going to announce it.
Model retirement is the most underrated. OpenAI's retirement notices are clear and dated, but they land in a changelog rather than in the conversation that breaks. The February 2026 round alone removed GPT-4o, GPT-4.1, GPT-4.1 mini and o4-mini from ChatGPT; o3 goes on August 26, 2026 following a 90-day sunset. Each notice adds that the change applies to ChatGPT only, with no equivalent change in the API — so the same model can be "gone" in one product and live in the other.
Allowances produce the most confusing symptom, because the app keeps working. OpenAI's Pro article is unusually direct about the consequences: reaching a model's allowance "does not by itself mean that your account was restricted", there is no setting to increase or bypass it, and support does not reset limits. If a model is still unavailable after the reset time shown in-product, that is the point at which it becomes a genuine fault worth reporting.
Lockdown Mode deserves its place because it is new, opt-in, and silently comprehensive. It restricts live web browsing, deep research, agent mode, file downloads and some web-derived image support, in order to reduce data-exfiltration risk from prompt injection. Every one of those, from the user's chair, looks like a feature that has stopped working.
Symptom → cause.
Almost never deletion. OpenAI's own checklist, in order: confirm you are signed into the correct account or workspace including the right SSO method; check Archived Chats in Settings → Data controls, because archiving hides conversations without deleting them; confirm the chat history toggle is on; refresh or sign out and back in; and check the status page, since an incident can hide history.
If you want to know what is genuinely still stored, request a data export from Settings → Data controls — that is the authoritative answer rather than what the sidebar happens to render.
The exact message — "You do not have an account because it has been deleted or deactivated" — means what it says, and the two halves behave differently. A deleted account is gone permanently; you can register the same email afresh after 30 days. A deactivated account does not come back by waiting: OpenAI states plainly that waiting 30 days does not reactivate one, and points you to check your email for instructions about deactivation, age verification or a policy issue.
If you did not delete it yourself, that email is the thing to find before contacting support.
Three candidates, in the order worth checking:
- Allowance spent — image generation, uploads and data analysis each meter separately. The reset time appears in-product.
- Lockdown Mode — if browsing, deep research, agent mode or downloads are the dead features specifically, this is the likely cause. Settings → Security on a personal account.
- Plan limits — some features simply are not included at your tier. Go excludes GPT-5.6 Sol and legacy models; Sites and Work are not on Free or Go; GPT creation is no longer on any personal plan at all.
The third is not a fault, and no amount of troubleshooting will produce the feature. Our page on ChatGPT's limits sets out which is which.
Two structural changes explain most of these, and neither is a bug. First, retirement: a GPT built around a model that has since been removed from ChatGPT will not behave as it did. OpenAI's notices confirm existing conversations continue on a successor model, which is exactly why the output changes.
Second, and more permanent: new GPT creation and publishing are no longer available on personal accounts — Free, Go, Plus and Pro alike — and now live in Business, Enterprise and Edu workspaces. Existing GPTs still run, and editing one depends on your plan and permissions. If the "create" option has vanished rather than errored, that is why.
Somewhere to work in the meantime.
If the status page is red, the fastest fix is a different vendor for an hour — every one of these has a free tier and none requires a card. We have not benchmarked quality between them; this is about having somewhere to go.
Frequently asked.
Quick follow-ups people search after this question.